Author + information
- Received June 17, 2015
- Revision received July 9, 2015
- Accepted July 10, 2015
- Published online December 1, 2015.
- Jennifer A. Schaub, MD∗,
- Steven G. Coca, DO, MS†,
- Dennis G. Moledina, MBBS∗,
- Mark Gentry, MA, MLS‡,
- Jeffrey M. Testani, MD, MS∗ and
- Chirag R. Parikh, MD, PhD∗∗ ()
- ∗Department of Internal Medicine and Applied Translational Research, Yale University School of Medicine, New Haven, Connecticut
- †Ichan School of Medicine at Mount Sinai, New York, New York
- ‡John Cushing/John Hay Whitney Medical Library, Yale University School of Medicine, New Haven
- ↵∗Reprint requests and correspondence to:
Dr. Chirag R. Parikh, Section of Nephrology, Yale University School of Medicine, 60 Temple Street, Suite 6C, New Haven, Connecticut 06510.
Objectives This study sought to determine if amino-terminal pro-B-type natriuretic peptide (NT-proBNP) has different diagnostic and prognostic utility in patients with renal dysfunction.
Background Patients with renal dysfunction have higher NT-proBNP, which may complicate interpretation for diagnosis of acute decompensated heart failure (ADHF) or prognosis.
Methods We searched MEDLINE and EMBASE through August 2014 for studies with a subgroup analysis by renal function of the diagnostic or prognostic ability of NT-proBNP.
Results For diagnosis, 9 studies were included with 4,287 patients and 1,325 ADHF events. Patients were mostly divided into subgroups with and without renal dysfunction by an estimated glomerular filtration rate of 60 ml/min/1.73 m2. In patients with renal dysfunction, the area under the curve (AUC) for NT-proBNP ranged from 0.66 to 0.89 with a median cutpoint of 1,980 pg/ml, while the AUC ranged from 0.72 to 0.95 with a cutpoint of 450 pg/ml in patients with preserved renal function. For prognosis, 30 studies with 32,203 patients were included, and mortality in patients with renal dysfunction (25.4%) was twice that of patients with preserved renal function (12.2%). The unadjusted pooled risk ratio for NT-proBNP and mortality was 3.01 (95% confidence interval [CI]: 2.53 to 3.58) in patients with preserved renal function and was similar in patients with renal dysfunction (3.25; 95% CI: 2.45 to 4.30). Upon meta-regression, heterogeneity was partially explained if patients with heart failure or coronary artery disease were enrolled.
Conclusions NT-proBNP retains utility for diagnosis of ADHF in patients with renal dysfunction with higher cutpoints. Elevated NT-proBNP confers a worse prognosis regardless of renal function.
The measurement of amino-terminal pro-B-type natriuretic peptide (NT-proBNP) is an accepted diagnostic strategy to clarify whether a patient presenting with dyspnea has acute decompensated heart failure (ADHF) and is a useful prognostic marker for mortality in a broad spectrum of patients (1–3). Incorporation of NT-proBNP in clinical management decreases the cost of medical care and improves clinical outcomes (4–6).
B-type natriuretic peptides (BNPs) are a family of proteins secreted by the ventricles during ventricular wall stretch and are cleaved into a biologically active fragment, BNP, and a biologically inactive fragment, NT-proBNP. It is known that BNP increases natriuresis, diuresis. and vasodilation, although there is uncertainty regarding processing and clearance of natriuretic peptides (7). Important steps such as posttranslational glycosylation and cleavage by the proteases, furin and corin, are likely deranged in heart failure (HF) (8–10). Although it is commonly believed that NT-proBNP is more reliant on renal excretion than BNP, there is evidence that the kidneys clear both hormones equally (11). Furthermore, mass spectroscopy data have shown that commercial BNP and NT-proBNP assays measure the target peptides as well as precursor molecules and breakdown products (12).
Important issues remain about how to use clinical NT-proBNP assays appropriately in the clinical setting. Because more than 50% of patients with a diagnosis of ADHF suffer from decreased glomerular filtration rate (GFR), physicians encounter this clinical conundrum frequently (13,14). It would be clinically valuable to clarify whether commercially available NT-proBNP assays are diagnostically and prognostically useful in patients with renal insufficiency.
We sought to establish the performance of NT-proBNP for diagnosis of ADHF and prognosis in patients with renal dysfunction by undertaking a systematic review and meta-analysis of the literature. Our study consists of 2 separate questions. First, does renal dysfunction alter the diagnostic ability of NT-proBNP to detect ADHF? Second, does renal dysfunction alter the prognostic ability of NT-proBNP?
This study was performed in accordance with published guidelines for systematic review and meta-analysis of observational studies (15).
A study was eligible for inclusion if it included at least 50 patients >18 years of age. For our first question, concerning if renal dysfunction impacts the diagnostic ability of NT-proBNP for ADHF, a study was eligible if the outcome was diagnosis of ADHF, and included a subgroup analysis by kidney function (either by serum creatinine, serum cystatin C, estimated glomerular filtration rate [eGFR] or measured GFR). For our second question, concerning if renal dysfunction impacts the prognostic ability of NT-proBNP, a study was eligible if the outcome was either all-cause mortality, cardiac mortality, or major adverse cardiac events and included a subgroup analysis by kidney function. For both questions, we searched for acute and chronic renal insufficiency.
Data Sources and Searches
We performed 2 separate searches in Ovid Medline and Ovid Embase from inception (1946 and 1974, respectively) until August 2014 based on our 2 questions. An example of the full search criteria is included in the supplementary materials (Online Table 1). We additionally searched the Science Citation Index Expanded on Web of Science and reviewed the references of all selected studies. There was no language restriction.
For our first and second question, area under the curve (AUC), sensitivity, specificity, and positive or negative predictive value were acceptable outcome measures. We additionally included studies that reported hazard ratio, odds ratio, or relative risk (RR) for our second question.
Data Extraction and Quality Assessment
Two reviewers decided if studies were eligible for inclusion (J.A.S. and S.G.C.). Two reviewers (J.A.S. and D.G.M.) independently extracted information using a standardized data collection sheet and final data were decided by consensus. Because multiple studies did not provide complete data, we contacted the authors to complete a data sheet of summary information. For studies concerning diagnosis, quality was assessed by the QUADRAS criteria (16); for prognosis, quality was assessed by the Hayden criteria (17).
Data Synthesis and Analysis
For the analysis regarding effect modification of renal function on diagnostic ability, we used the method developed by Moses et al. (18) to produce separate summary receiver operator curves (ROC) for patients with preserved and diminished renal function. This method does not allow estimation of accurate confidence intervals or summary AUC; thus, we examined whether there was effect modification by comparing the summary ROCs visually.
For assessing effect modification of renal function on prognostic value, we conducted separate analyses for patients with preserved and diminished renal function, and evaluated the difference between the pooled risk ratios with a 2-sample Z test. Crude and adjusted effect estimates were pooled separately to calculate risk ratios using techniques that accounted for within- and between-study heterogeneity (random effects method of DerSimonian and Laird) and were weighted by their inverse variance. If a study provided both a crude and adjusted effect estimate, it was included in both types of analysis. We formally assessed heterogeneity of effects between studies with the Cochran I and τ2 statistics.
If significant heterogeneity was present as indicated by the τ2 statistic, we undertook meta-regression to evaluate for sources of heterogeneity. We used a mixed effects linear regression model where the random effects were estimated the DerSimonian and Laird estimator and studies were weighted by the inverse variance method. Moderators were prespecified and were if the study population included patients with CAD versus other comorbidities, patients with HF versus other comorbidities, inpatients versus outpatients, renal function estimate (eGFR vs. other/serum cystatin C versus other), NT-proBNP cutpoints (same cutpoints for patients with preserved and diminished renal function versus different cutpoints), and NT-proBNP assay (Roche vs. other). Because there were few studies, each moderator was analyzed individually and we applied a Bonferroni correction factor to create a study wide alpha level of 0.05.
Meta-analysis of diagnostic data and meta-regression were conducted in RevMan version 3.1.2 (19,20). The remainder of the analysis was conducted in RevMan version 5.3.
We identified 4,012 studies eligible for review after excluding 657 duplicate citations. After title and abstract screening, we identified 236 articles for full text review. Nine articles were eligible to evaluate diagnostic performance and 30 articles were eligible to evaluate prognostic performance (Online Figure 1). Six studies for diagnosis and 16 studies for prognosis required further information from the authors, and we received responses from 5 authors. After contacting the authors, 5 studies for diagnostic performance and 17 studies on prognostic ability had complete quantitative data sufficient for meta-analysis.
Study Characteristics Regarding Diagnosis
The study and patient characteristics for diagnosis of ADHF are depicted in Table 1 and the summary data are presented in Table 2. The majority of the studies were conducted in the emergency department setting (21–25). Two studies used the standard “Januzzi cutoffs,” which indicates a NT-proBNP level of >450 pg/ml in patients <50 years of age and >900 pg/ml in patients >50 years of age (21,24); the remaining studies reported the optimal cutpoint from an ROC analysis from their cohort.
Study Characteristics Regarding Prognosis
The study and patient characteristics for prognosis are depicted in Table 3 and the summary data for studies that provided quantitative data by subgroup are presented in Table 4 for studies pertaining to prognosis. Follow-up for events was evaluated over a range of time intervals in the studies, spanning from 2 months to 10 years. The studies enrolled patients with various medical comorbidities, although the most common were HF and CAD. In more than one-half of the studies, the absolute event rates were 2-fold higher in patients with renal dysfunction than patients without renal dysfunction.
Correlation of GFR and NT-proBNP Levels
Eleven studies reported correlation coefficients between eGFR and NT-proBNP (21,22,24,26–33). The correlation coefficients were statistically significant and ranged from -0.21 to -0.58. A subset of studies reported NT-proBNP levels by patient eGFR; NT-proBNP levels consistently increase as renal function declines (Online Table 2).
Diagnostic Ability of NT-proBNP
For diagnosing ADHF, the cutpoints in patients with an eGFR of <60 ml/min/1.73 m2 were roughly 2-fold higher than the cutpoints in patients with an eGFR of >60 ml/min/1.73 m2 (Table 2). Even with higher cutpoints, the specificity and sensitivity were often slightly lower in patients with an eGFR of <60 compared with patients with an eGFR of >60 (Figure 1A). Although we were unable to generate confidence intervals for the AUC to evaluate if the summary ROCs are significantly different, the curves overlap on visual inspection (Figure 1B).
Prognostic Ability of NT-proBNP
Twelve studies provided crude estimates for the association between NT-proBNP and mortality, either in the form of event rates, RR or sensitivity/specificity data. When comparing patients with preserved renal function and elevated NT-proBNP with patients with preserved renal function and normal NT-proBNP, the unadjusted pooled RR was 3.01 (95% CI: 2.53 to 3.58) (Figure 2A). Correspondingly, the pooled risk ratio for patients with renal dysfunction was 3.25 (95% CI: 2.45 to 4.30), although there was a higher event rate (20.9%) than patients with preserved renal function (11.9%) (Figure 2B). The pooled risk ratios between patients with preserved and diminished renal function were not different (p = 0.652), and there was significant heterogeneity in both subgroups (Figures 2A and 2B). Four of the studies used higher NT-proBNP cutpoints for patients with renal dysfunction (28,33–35), whereas the remaining studies used the same NT-proBNP cutpoints in the different subgroups of renal function.
Nine studies provided adjusted estimates for the association between NT-proBNP and mortality. Although the studies adjusted for different variables, the pooled risk ratios for patients with preserved and diminished renal function were similar (RR: 1.43, 95% CI: 1.34 to 1.53 and RR: 1.59, 95% CI: 1.41 to 1.80, respectively) (Figures 3A and 3B), and there was no significant difference between the 2 groups (p = 0.141). There was significant heterogeneity in patients with renal dysfunction, which resolved after removing the study by Waldum et al. (36) (Online Figure 2). The authors noted in correspondence that these data did not meet the proportional hazards assumption.
Assessment of Bias
Funnel plots were created to assess publication bias (Online Figure 3). Funnel plots for crude effect estimates demonstrated heterogeneity but were not markedly asymmetric. There was publication bias favoring small studies with large adjusted effect estimates, especially in patients with renal dysfunction.
Meta-Regression for Crude Effect Estimates of NT-proBNP and Mortality
Because there was significant heterogeneity in the crude effect estimates for NT-proBNP and mortality, we undertook meta-regression. Samples comprised of patients with CAD or HF explained significant proportions of the heterogeneity (Online Table 3). In patients with renal dysfunction, the slope was significantly positive if studies sampled patients with CAD suggesting that increased NT-proBNP confers a greater magnitude increase in risk of mortality in patients with renal dysfunction and CAD compared with patients with renal dysfunction and other comorbidities. In patients with preserved or diminished renal function, the slope was significantly negative if studies sampled patients with congestive HF, suggesting that increased NT-proBNP confers a lesser magnitude increase in risk of mortality in patients with HF compared with patients with other comorbidities. The type of renal function estimate, presence of different NT-proBNP cutpoints for kidney function subgroups, or NT-proBNP assay failed to explain a significant proportion of the heterogeneity.
Although patients with renal dysfunction have higher plasma levels of NT-proBNP, this systematic review and meta-analysis demonstrates that NT-proBNP remains useful for diagnosis of ADHF or prognosis in these patients. If higher cutpoints are used, NT-proBNP still has acceptable sensitivity and specificity for diagnosis of ADHF in patients with renal dysfunction. Furthermore, increased NT-proBNP in patients with renal dysfunction compared with patients with normal NT-proBNP and renal dysfunction confers similarly increased risk of mortality as in patients with preserved renal function. This relationship was consistent among different patient care settings, although meta-regression found that the relationship between NT-proBNP and mortality varied depending on the patients’ comorbidities.
Diagnosis of ADHF
Among the studies identified addressing the diagnostic ability of NT-proBNP for ADHF, the cutpoints varied from 1,200 to >6,000 pg/ml for patients with an eGFR of <60 ml/min/1.73 m2, all of which are higher than the standard “Januzzi cutpoints.” Despite the higher cutpoints, the AUC, sensitivity, and specificity in patients with renal dysfunction was often slightly lower, although the summary ROC curves for patients with preserved and diminished renal function overlap on visual inspection. Higher NT-proBNP cutpoints did not always result in greater sensitivity and lesser specificity; this may be because the “gold standard” was different in each study. Furthermore, the “gold standard” often partially entailed chart review by a clinician, which relies heavily on an individual’s clinical judgment and likely varied between studies. Thus, although it seems that NT-proBNP retains diagnostic utility if using higher cutpoints in patients with renal dysfunction, it is possible that more patients would be identified falsely with ADHF in clinical practice. We would encourage prospective clinical studies be completed specifically for patients with renal dysfunction to help identify more precise cutpoints.
Regarding prognosis, the crude and adjusted pooled RRs for NT-proBNP were not different between patients with preserved and diminished renal function, although the absolute event rates were 2-fold greater in patients with diminished renal function. This suggests that, among patients with renal dysfunction, increased NT-proBNP confers the same increased risk of mortality as compared with increased NT-proBNP in patients without renal dysfunction. However, because patients with renal dysfunction are overall at higher risk of mortality, the absolute mortality rates are higher than patients without renal dysfunction. Thus, increased NT-proBNP levels in patients with renal dysfunction are still capturing important prognostic information. Because it is uncertain if NT-proBNP is completely renally excreted, markedly increased levels of NT-proBNP in patients with diminished renal function could be due to either decreased clearance or increased cardiac production. Both of these factors could contribute to increased mortality.
There was significant heterogeneity for the pooled unadjusted risk ratios, although the adjusted risk ratios did not have significant heterogeneity after excluding 1 study. Underlying patient population comorbidities were a significant contributor to heterogeneity in meta-regression, which suggests that the prognostic value of NT-proBNP may vary depending on the patient’s clinical diagnosis. It seems that increased NT-proBNP confers a greater magnitude increase in mortality in patients with renal dysfunction and CAD compared with increased NT-proBNP in patients with HF. This is likely because increased NT-proBNP in patients with CAD provides new information that the patient additionally has ventricular stress, whereas increased NT-proBNP in patients with HF does not provide new information regarding the presence of ventricular stress. It was reassuring that the type of renal function estimate, NT-proBNP assay, or different NT-proBNP cutpoints for kidney function subgroups did not contribute significantly to the observed heterogeneity.
As the studies in this systematic review demonstrate, NT-proBNP levels are often markedly increased in patients with renal dysfunction, which presents a conundrum in clinical practice. Is this cardiac biomarker elevated because of diminished renal clearance or because of extrarenal pathophysiology? The specific mechanisms for clearance of NT-proBNP are not fully elucidated, so these results could be due to either decreased clearance or increased cardiac production. It is plausible that these results are due to increased cardiac production since the anemia, uremia and secondary hyperparathyroidism from chronic kidney disease is known to have myriad negative ramifications on myocardium (37–39).
First, the studies varied regarding NT-proBNP assay, NT-proBNP cutpoints, renal function measurements, and how patients were divided into subgroups. This could introduce heterogeneity into our study, although this was not detected by meta-regression techniques. Second, there were few studies regarding diagnosis of ADHF and, hence, we were unable to analyze the data with statistically rigorous methodology to determine whether there was a significant difference between the summary ROCs. Third, we did not have patient-level data and were unable to identify optimal cutpoints for NT-proBNP in diagnostic and prognostic scenarios. Fourth, there were few patients with end-stage renal disease in these studies, so these results should not be extrapolated to the dialysis population. Fifth, many patients with HF have dynamic renal function, thus the NT-proBNP cutpoints in the diminished renal function subgroups may not be precise.
This meta-analysis demonstrates that assessment of NT-proBNP levels in patients with preserved and diminished renal function provides useful prognostic information regarding mortality and would likely have diagnostic utility if cutpoints were identified for various GFR categories. Future studies should explore establishing accepted diagnostic cutpoints for ADHF in patients with diminished GFR. Although markedly increased NT-proBNP in a patient with renal dysfunction may be partially due to decreased clearance, it still portends a higher absolute risk for mortality compared with patients with normal renal function. It would be beneficial to elucidate how natriuretic peptide processing is altered in patients with diminished renal function. There is a relative dearth of information regarding the test characteristics of novel biomarkers in patients with renal dysfunction. Investigators should make every effort to sample a sufficient number of patients with renal disease and investigate the test characteristics separately in this important and vulnerable group of patients.
COMPETENCY IN MEDICAL KNOWLEDGE: NT-proBNP levels are often increased in patients with renal dysfunction, which may complicate interpretation for diagnosis of ADHF or prognosis. For the diagnosis of ADHF, NT-proBNP seems to have similar diagnostic properties in patients with renal dysfunction compared with patients without renal dysfunction provided higher cutoffs are used. Although patients with renal dysfunction have an absolute mortality rate that is approximately double that of patients without renal dysfunction, NT-proBNP still captures useful prognostic information.
TRANSLATIONAL OUTLOOK: NT-proBNP levels are commonly increased in patients with renal dysfunction, although it is not clear if this is due to increased cardiac production or decreased renal clearance. Future studies should clarify the biology of NT-proBNP production and clearance, especially in patients with renal dysfunction. Investigators should make a concerted effort to separately analyze biomarker characteristics in patients with renal dysfunction.
The authors thank Drs. Fabbian, Lu, Darmon, Waldum, and Masson (on behalf of the Val-HeFT trial) who kindly provided further data for the meta-analysis.
Dr. Schaub is supported by NIH (T32DK007276-36). Dr. Testani is supported by NIH (K23HL114868 and L30HL115790). Dr. Parikh is supported by the NIH (K24DK090203).
All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- acute decompensated heart failure
- area under the curve
- B-type natriuretic peptide
- coronary artery disease
- glomerular filtration rate
- estimated glomerular filtration rate
- heart failure
- plasma amino amino-terminal pro-B-type natriuretic peptide
- receiver operator curves
- relative risk
- Received June 17, 2015.
- Revision received July 9, 2015.
- Accepted July 10, 2015.
- American College of Cardiology Foundation
- Hartmann F.,
- Packer M.,
- Coats A.J.,
- et al.
- Jessup M.,
- Abraham W.T.,
- Casey D.E.,
- et al.
- Laramee P.,
- Wonderling D.,
- Swain S.,
- Al-Mohammad A.,
- Mant J.
- Moe G.W.,
- Howlett J.,
- Januzzi J.L.,
- Zowall H.
- Huntley B.K.,
- Sandberg S.M.,
- Heublein D.M.,
- Sangaralingham S.J.,
- Burnett J.C. Jr..,
- Ichiki T.
- Semenov A.G.,
- Tamm N.N.,
- Seferian K.R.,
- et al.
- Vodovar N.,
- Seronde M.F.,
- Laribi S.,
- et al.
- van Kimmenade R.R.,
- Januzzi J.L. Jr..,
- Bakker J.A.,
- et al.
- Miller W.L.,
- Phelps M.A.,
- Wood C.M.,
- et al.
- Heywood J.T.,
- Fonarow G.C.,
- Costanzo M.R.,
- Mathur V.S.,
- Wigneswaran J.R.,
- Wynne J.
- Smith G.L.,
- Lichtman J.H.,
- Bracken M.B.,
- et al.
- Doebler P, Holling H. Meta-analysis of diagnostic accuracy with mada. 2013. Available at: https://cran.rproject.org/web/packages/mada/vignettes/mada.pdf. Accessed November 24, 2014.
- Anwaruddin S.,
- Lloyd-Jones D.M.,
- Baggish A.,
- et al.
- Chenevier-Gobeaux C.,
- Claessens Y.E.,
- Voyer S.,
- Desmoulins D.,
- Ekindjian O.G.
- DeFilippi C.R.,
- Seliger S.L.,
- Maynard S.,
- Christenson R.H.
- Astor B.C.,
- Yi S.,
- Hiremath L.,
- et al.
- Coquet I.,
- Darmon M.,
- Doise J.M.,
- et al.
- Fu S.,
- Luo L.,
- Ye P.,
- et al.
- van Kimmenade R.R.,
- Januzzi J.L. Jr..,
- Baggish A.L.,
- et al.
- Masson S.,
- Latini R.,
- Anand I.S.,
- et al.
- Waldum B.,
- Stubnova V.,
- Westheim A.S.,
- Omland T.,
- Grundtvig M.,
- Os I.
- Lekawanvijit S.,
- Adrahtas A.,
- Kelly D.J.,
- Kompa A.R.,
- Wang B.H.,
- Krum H.
- London G.